An introduction to physics-informed neural networks (PINNs), which integrate physical laws expressed as PDEs into neural network training by enforcing differential equation constraints through automatic differentiation.

A short overview of my MSc thesis, which adapts and applies Sequential and Prior Likelihood Mixing (Kirschner et al., 2025) to anytime-valid uncertainty quantification in tomographic imaging. The work explores how to obtain statistically valid uncertainty for SPECT reconstructions using both classical estimators and modern neural predictors such as U-Net ensembles and diffusion models.

This post introduces mini-mcmc, a compact Rust library for running modern MCMC algorithms—like NUTS, HMC, Metropolis–Hastings, and Gibbs—with automatic differentiation, parallel chains, and optional GPU acceleration.

A quick introduction to the basics of conformal prediction under exchangeability.